replace all these cheap relays/optoMOS by more adequate models
well, don't care that much about CAT II specs, 10x probe is standard for me.
replace BAV99 diodes
why? They good enough.
replace the 2 BC846 transistor for current mirror by at least a matched pair (BC846B)
that's already matched pair of BC846B (actually you don't have to much them today anymore)
replace the AD8310/LMH6552/VariCap by the LMH6518 above
for what? Actually you mean AD8370 and not 8310 (or you mean the AD8510 in LF path?)
Anyway, there is no reason to replace them. I can tell you something, Rigol did it in 300MHz CA models
and switched back to AD8370/LMH6552 because this combination works better.
Varicap is actually only part of the story, it 20MHz bw limiter, yes, but it does have only
mariginal influence on HF bw (just calculate back the filter corner freq. given by varicap+caps, out/in imp. and the
resistor networks and you will see it is much higher than necessary. The bw filter is within firmware)
straighten the signal trace as much as possible
hmm, it is stight, then 45° to first opamp then again stright ... the only think i don't like
it are the replays switching ADCs together, but the influence is as far as possible reduced
by self-calibration
keep the signal trace as short as possible and away from other traces
that would be complettly new PCB, much bigger anyway. Hantek/Tekway did it anyway better
(they have at least better shielding) than Rigol/ATTEN where the PSU is so close to ADCs/Trigger
stage/1MSs-500MSs relays that you have already influence.