Try it with WHDLoad v16, which was current at the time the slave was written. Despite the claimed performance tweaks in the readme, I suspect there's a bug hindering performance, possibly triggered by changes in newer WHDLoad versions.
Also try the CACHE tooltype to force-enable that feature. I was also going to suggest the MMU tooltype (which is related to cache performance) but the
docs say the external MMU on the 2620 (I assume that's your accelerator judging by the boot menu) isn't supported. But try it anyway.
