Erik wrote:
"you will still want your "heavy use" variables in (I)DATA for speed."

Also for code size; accessing (I)DATA takes less code than accessing PDATA, and a lot less code than accessing XDATA!

As I've mentioned before (possibly on the Keil forum), I recently worked on a Client Project where moving just one variable from XDATA to DATA gave several K-byte reduction in code size!

Note that you can use the Keil uVision code browser to identify your most-frequently-used variables - these are prime targets for (I)DATA...!