after playing around with my own linux system, it appears that the read performance of this box could be improved by some tuning of buffer sizes and tcp options. right now, a 4k jumbo packet 'read' operation on this box is no faster than a normal 1500 mtu operation.
perhaps the ibm redbook on linux performance tuning could be mentioned to your slaves who are doing the firmware work.
<update>
at this point it seems the transfer rates are on par with similar and even more expensive units, so I am moderating my stand on this. however, one must ask if jumbo packet size is a true feature as it provides no benefits. if this is a hardware horsepower limitation, why bother 'supporting' jumbo packets? or if you are going to support jumbo packets, why not at least find out the true limiting factor? is it the processor speed? lack of dma? not enough buffers? tcp tuning?
it seems even a 'atom' based tiny pc can provide transfer rates in the area of 2-3x that these cute little boxes can...