Performance of Windows Phone 7 apps are often vastly different on the emulator than on actual device. Many things - I/O, networking, libraries, floating point math etc - are even different from what we're used to for full desktop .NET apps. For instance, on some devices floating point math is much faster than integer math, unlike for desktop apps. That makes it rather hard to guess your WP7 app's performance bottlenecks, and some "optimizations" can actually slow things down on real device.
EQATEC brings a code-profiler that can profile WP7 Silverlight app on real phone hardware, as well as on emulator. The profiler is free for single-assembly apps, and also does full desktop, NETCF, and Silverlight profiling.
This short video show how a few apps were profiled on an actual WP7 phone on loan from Microsoft Denmark.