Jeffrey Wang wrote a very detailed blog post about recursive calculations in MDX with Analysis Services. In these days I’m doing some comparison between DAX and MDX (well, between Vertipaq/PowerPivot and OLAP/SSAS engines) and I’m discovering that MDX calculation engine is much more smart than you might think in certain conditions. I cannot anticipate very much (also because I’m far from conclusions by now), but it’s clear that the combination of bulk mode evaluation and the several types of caches that are used by the OLAP engine can perform really well. As Jeffrey shows in his post, the problem is that bulk-mode evaluation is very important and its usage is very sensitive to the way the MDX query is built. You can fix the MDX query in order to improve performance, but this might be very hard whenever the query is automatically generated by a cube browser, like Excel. In these cases, Vertipaq/PowerPivot seems to offer a more predictable response time (not always faster than OLAP, anyway!).
Now back to study, I hope we’ll discuss much more about these things when Denali will be publicly available with CTP3!