jffs2 write behavior

[Agarwal, Anil]

I am using jffs2 on an embedded powerpc system with 64KB-sector NOR
flash.
 
If I write a single file, 128 bytes at a time, using write(), then
the time taken by each sequential write operation is generally as
follows -
 
0. t 30 ms
1. t 50 ms
2. t 68 ms
3. t 86 ms
4. t 105 ms
5. t 123 ms
6. t 142 ms
7. t 159 ms
8. t 177 ms
9. t 195 ms
10. t 213 ms
11. t 231 ms
12. t 248 ms
13. t 267 ms
14. t 285 ms
15. t 302 ms
16. t 321 ms
17. t 349 ms
18. t 358 ms
19. t 374 ms
20. t 393 ms
21. t 410 ms
22. t 428 ms
23. t 446 ms
24. t 464 ms
25. t 481 ms
26. t 503 ms
27. t 577 ms
28. t 534 ms
29. t 552 ms
30. t 570 ms
31. t 588 ms

32. t 32 ms
33. t 50 ms
34. t 68 ms
35. t 87 ms
36. t 104 ms
37. t 123 ms
38. t 141 ms
39. t 159 ms
40. t 177 ms
41. t 195 ms
42. t 213 ms
43. t 231 ms
...
 
It almost appears that jffs2 writes 128 bytes to flash the first time,
256 bytes the second time, 
and so on until it reaches 4096 bytes (the Linux page size?).
The flash write time for 128 bytes is ~20 ms, based on the chip spec.
 
If the above observation is true, then it would imply that
jffs2 wrote more than 64 KBytes to flash for 4 KBytes of real file data.
 
Of course, one can use buffered writes (fwrite) to improve this,
but the above behavior seems suspicious.
Is it a consequence of writing out page buffers from the start of the
page buffer
in jffs2_write_commit()?
 
Can someone please clarify if this is well-known/expected or
if there is something wrong with our jffs2 configuration? 
 
Thanks,
Anil Agarwal