You can use Event Tracing for Windows (ETW) to trace your extension DLL execution. ETW has almost no overhead when no listener is active, so in normal conditions your DLL will incur no perf penalty; at the same time it allows for detailed output at various levels of details.
The way ETW works is when the APIs are called, they check if there is a listener subscribed to the traces from particular publisher and if no, nothing is generated. If there is a listener, only the traces to which the listener is subscribed are written to a memory-mapped file. Thus, only as much traces data is generated as requested.
ETW listeners can be activated at any time and the publisher does not have to be restarted. Also, ETW is not flavor bound and can be used in both debug and retail. Thus, if a customer of yours has a problem, you have to only send them the listener with instructions on how to run it and collect the info; you don't have to sent them an instrumented binary version. You can either write your own app that acts as a listener, or you can use the standard tracelog.exe and tracefmt.exe tools to get the traces written to a file.
To generate the necessary ETW code in your DLL, you can use the WPP preprocesor instead of directly using the ETW APIs.
Note: While all the links I post here are to the Windows Driver Kit documentation, ETW and WPP can be (and are heavily) used for regular user mode programs.