Hi mlefevre,
The performance you are mentioning is normal. Microblaze is a slow processor, but performance can be increased with:
Caching
Higher frequency
Processor configuration (area optimized significantly reduces performance)
Memory controller and AXI bus width
The real bandwidth is in FPGA logic. For example to move data you implement DMA and to process it you need some parallel processing core. The AXI_DMA can achieve 900MB/s of sustainable bandwidth at 128bits@62MHz (tested on one memory bank of GigaBee). With parallel processing in the FPGA logic you can achieve unbelievable performance if compared to sequential processing (compared to PC).
But this can only be achieved using RTL programs (this is not C code!).
Best regards,
Ales