just a little more info, after changing the delay(10000) to delay(1000)
in esp.c, a profiled kernel now shows:

...
Each sample counts as 0.0166667 seconds.
  %   cumulative   self              self     total
 time   seconds   seconds    calls  us/call  us/call  name
 47.48    586.07   586.07                             mcount
 23.12    871.48   285.41    29201  9773.92 12858.86  esp_quick_dma_go
 10.97   1006.84   135.37                             Idle
  7.29   1096.83    89.98 119295919     0.75     0.75  esp_dafb_have_dreq
  2.47   1127.31    30.48    22857  1333.65  1333.65  copyout
  2.00   1151.97    24.67    58792   419.56   419.56  copyin
  1.27   1167.62    15.65    57540   271.98   271.98  delay
...

this looks fairly 'right' to me.  esp_quick_dma_go seems to dominate
and delay is now down in the noise region.  it also seems to point out
that esp_dafb_have_dreq should be quick quick quick.