This guy is correct, and to expand a bit further: wafflepool (in an attempt to make our servers cry less) set our max miner difficulty pretty high, so variance goes up for shares submitted, but overall profits stay the same. Less load on our servers, slightly less accurate mining speed stats.
One idea would be to scale the hash rate calculation by shares. Instead measuring shares in the last so many minutes, measure minutes for the last N shares (with N chosen to give a reasonably accurate hash rate estimate). This will give slower hash rate estimates for small miners, but more accurate estimates. Or likewise some similar exponential moving average which might be easier to compute.
Definitely an option. If we have enough complaints/confusion we'll look into switching it. Would just take some database index changes (everything is optimized for time-based currently)
