I agree ... I benchmarked tempdb with different configurations - tempdb on disk with a good cache exclusively allocated to it seems as good as in-memory (in our configuration). I wouldn't assume everyone will get the same results. The last benchmark I tried on different hardware showed in memory was 10% faster than on disk,
We have multiple tempdb's which need to be quite large for the occasional times in the week where we need lots of tempdb. However, for 99% of the time we only need a quarter of the tempdb - so the tempdb cache is a quarter the size it would be if it was in-memory leaving more space for real data.