POI或者JXL在导出大量数据的时候，由于它们将每一个单元格生都成一个Cell对象，所以很容易导致内存溢出。解决这个问题，唯一的办法是弄清楚Excel的二进制格式（汗），并且用流的方式读写Excel。POI和JXL其实提供了二进制方式读写Excel的API，只是因为缺少文档和实例，所以使用的人不多。我编写了这个简单的合并Excel的类，它只适合合并结构相同的多个Excel文件。好在这个功能已经可以解决数据导出产生OOM的问题：将数据分批导出然后合并。
下面的代码使用POI3.1，合并11个3000多行的文档用时约6秒，我实在找不到更多的测试用的文档了。
 

Java代码
@SuppressWarnings("unchecked")  
public class XlsMergeUtil {  
  private static Logger logger = LoggerFactory.getLogger(XlsMergeUtil.class);  

/** 
 * 将多个Xls文件合并为一个，适用于只有一个sheet，并且格式相同的文档 
 * @param inputs 输入的Xls文件 
 * @param out 输出文件 
 */ 
 public static void merge(InputStream[] inputs, OutputStream out) {  
   if (inputs == null || inputs.length <= 1) {  
     throw new IllegalArgumentException("没有传入输入流数组,或只有一个输入流.");  
   }  
 
   List<Record> rootRecords = getRecords(inputs[0]);  
   Workbook workbook = Workbook.createWorkbook(rootRecords);  
   List<Sheet> sheets = getSheets(workbook, rootRecords);  
   if(sheets == null || sheets.size() == 0) {  
     throw new IllegalArgumentException("第一篇文档的格式错误，必须有至少一个sheet");  
   }  
   //以第一篇文档的最后一个sheet为根，以后的数据都追加在这个sheet后面  
   Sheet rootSheet = sheets.get(sheets.size() - 1);   
   int rootRows = getRowsOfSheet(rootSheet); //记录第一篇文档的行数，以后的行数在此基础上增加  
   rootSheet.setLoc(rootSheet.getDimsLoc());  
   Map<Integer, Integer> map = new HashMap(10000);  
 
   for (int i = 1; i < inputs.length; i++) { //从第二篇开始遍历  
     List<Record> records = getRecords(inputs[i]);  
     int rowsOfCurXls = 0;  
     //遍历当前文档的每一个record  
     for (Iterator itr = records.iterator(); itr.hasNext();) {  
       Record record = (Record) itr.next();  
       if (record.getSid() == RowRecord.sid) { //如果是RowRecord  
         RowRecord rowRecord = (RowRecord) record;  
         //调整行号  
         rowRecord.setRowNumber(rootRows + rowRecord.getRowNumber());  
         rootSheet.addRow(rowRecord); //追加Row  
         rowsOfCurXls++; //记录当前文档的行数  
       }  
       //SST记录，SST保存xls文件中唯一的String，各个String都是对应着SST记录的索引  
       else if (record.getSid() == SSTRecord.sid) {  
         SSTRecord sstRecord = (SSTRecord) record;  
         for (int j = 0; j < sstRecord.getNumUniqueStrings(); j++) {  
           int index = workbook.addSSTString(sstRecord.getString(j));  
           //记录原来的索引和现在的索引的对应关系  
           map.put(Integer.valueOf(j), Integer.valueOf(index));  
         }  
       } else if (record.getSid() == LabelSSTRecord.sid) {  
         LabelSSTRecord label = (LabelSSTRecord) record;  
         //调整SST索引的对应关系  
         label.setSSTIndex(map.get(Integer.valueOf(label.getSSTIndex())));  
       }  
       //追加ValueCell  
       if (record instanceof CellValueRecordInterface) {  
         CellValueRecordInterface cell = (CellValueRecordInterface) record;  
         int cellRow = cell.getRow() + rootRows;  
         cell.setRow(cellRow);  
         rootSheet.addValueRecord(cellRow, cell);  
       }  
     }  
     rootRows += rowsOfCurXls;  
   }  
   byte[] data = getBytes(workbook, sheets.toArray(new Sheet[0]));  
   write(out, data);  
 }  
 
 static void write(OutputStream out, byte[] data) {  
   POIFSFileSystem fs = new POIFSFileSystem();  
   // Write out the Workbook stream  
   try {  
     fs.createDocument(new ByteArrayInputStream(data), "Workbook");  
     fs.writeFilesystem(out);  
     out.flush();  
   } catch (IOException e) {  
     e.printStackTrace();  
   } finally {  
     try {  
       out.close();  
     } catch (IOException e) {  
       e.printStackTrace();  
     }  
   }  
 }  
 
 static List<Sheet> getSheets(Workbook workbook, List records) {  
   int recOffset = workbook.getNumRecords();  
   int sheetNum = 0;  
 
   // convert all LabelRecord records to LabelSSTRecord  
   convertLabelRecords(records, recOffset, workbook);  
   List<Sheet> sheets = new ArrayList();  
   while (recOffset < records.size()) {  
     Sheet sh = Sheet.createSheet(records, sheetNum++, recOffset);  
 
     recOffset = sh.getEofLoc() + 1;  
     if (recOffset == 1) {  
       break;  
     }  
     sheets.add(sh);  
    }  
    return sheets;  
  }  
 
  static int getRows(List<Record> records) {  
    int row = 0;  
    for (Iterator itr = records.iterator(); itr.hasNext();) {  
      Record record = (Record) itr.next();  
      if (record.getSid() == RowRecord.sid) {  
        row++;  
      }  
    }  
    return row;  
  }  
    
  static int getRowsOfSheet(Sheet sheet) {  
    int rows = 0;  
    sheet.setLoc(0);  
    while(sheet.getNextRow() != null) {  
      rows++;  
    }  
    return rows;  
  }  
 
  @SuppressWarnings("deprecation")  
  static List<Record> getRecords(InputStream input) {  
    try {  
      POIFSFileSystem poifs = new POIFSFileSystem(input);  
      InputStream stream = poifs.getRoot().createDocumentInputStream("Workbook");  
      return org.apache.poi.hssf.record.RecordFactory.createRecords(stream);  
    } catch (IOException e) {  
      logger.error("IO异常：{}", e.getMessage());  
      e.printStackTrace();  
    }  
    return Collections.EMPTY_LIST;  
  }  
 
  static void convertLabelRecords(List records, int offset, Workbook workbook) {  
 
    for (int k = offset; k < records.size(); k++) {  
      Record rec = (Record) records.get(k);  
 
      if (rec.getSid() == LabelRecord.sid) {  
        LabelRecord oldrec = (LabelRecord) rec;  
 
        records.remove(k);  
        LabelSSTRecord newrec = new LabelSSTRecord();  
        int stringid = workbook.addSSTString(new UnicodeString(oldrec.getValue()));  
 
        newrec.setRow(oldrec.getRow());  
        newrec.setColumn(oldrec.getColumn());  
        newrec.setXFIndex(oldrec.getXFIndex());  
        newrec.setSSTIndex(stringid);  
        records.add(k, newrec);  
      }  
    }  
  }  
 
  public static byte[] getBytes(Workbook workbook, Sheet[] sheets) {  
    // HSSFSheet[] sheets = getSheets();  
    int nSheets = sheets.length;  
 
    // before getting the workbook size we must tell the sheets that  
    // serialization is about to occur.  
    for (int i = 0; i < nSheets; i++) {  
      sheets[i].preSerialize();  
    }  
 
    int totalsize = workbook.getSize();  
 
    // pre-calculate all the sheet sizes and set BOF indexes  
    int[] estimatedSheetSizes = new int[nSheets];  
    for (int k = 0; k < nSheets; k++) {  
      workbook.setSheetBof(k, totalsize);  
      int sheetSize = sheets[k].getSize();  
      estimatedSheetSizes[k] = sheetSize;  
      totalsize += sheetSize;  
    }  
 
    byte[] retval = new byte[totalsize];  
    int pos = workbook.serialize(0, retval);  
 
    for (int k = 0; k < nSheets; k++) {  
      int serializedSize = sheets[k].serialize(pos, retval);  
      if (serializedSize != estimatedSheetSizes[k]) {  
            throw new IllegalStateException("Actual serialized sheet size (" + serializedSize  
            + ") differs from pre-calculated size (" + estimatedSheetSizes[k] + ") for sheet (" + k  
            + ")");  
        Sheet.serializeIndexRecord() does not  
      }  
      pos += serializedSize;  
    }  
    return retval;  
  }  
 
  public static void main(String[] args) throws Exception {  
    final String PATH = "E:\\projects\\java\\ws_0\\export\\data\\";  
    InputStream[] inputs = new InputStream[10];  
    inputs[0] = new java.io.FileInputStream(PATH + "07_10.xls");  
    for(int i = 1; i <= 9; i++) {  
      inputs[i] = new java.io.FileInputStream(PATH + "07_0" + i + ".xls");  
    }  
    OutputStream out = new FileOutputStream(PATH + "xx.xls");  
    long t1 = System.currentTimeMillis();  
    merge(inputs, out);  
    System.out.println(System.currentTimeMillis() - t1);//简陋的测试一下时间  
  }  
 
}