All of the classical big tools, such as Mercury/HP LoadRunner, Silk/Borland/Microfocus Performer & Rational/IBM Performance Tester all have strong client-server testing backgrounds. Most of these tools began in the two tier client-server arena and support collection (recording) or development in dozens (in once case over 30) difference interfaces.
LoadRunner historically has been the technically strongest tool in this set, supporting the largest number and types of interfaces, followed by Borland/Microfocus and IBM. There was a note above asking about architecture of the add ins and this is key. Understanding what is the next upstream server component and how the widget communicates is critical. You may even have to mix protocols/interfaces in a single piece of test code and you will need a tool which supports them all.