As you say, traffic shaping is an option and imho would be your best bet here.
You could also alter the MTU of the links, which in turn would increase fragmentation and reduce efficency (thus increasing latency).
Unfortunatly these techniques will only give approximated behaviour and will not produce hard and fast data (e.g. it'll be hard to simulate propogation delay if using disimilar physical mediums). If you need that level of assurance then building a lab that is closely matches the production system may be required.
The trouble you'll face in general is that unless you use a packet generator tuned to simulate your production network, any test you do in isolation will be based on disimilar operating conditions. In which case the best you can hope for is an approximation of behaviour.
Depending on whether you're testing the application itself or its impact on the network you could use a network analysis system (e.g. Smartbits - http://www.spirentcom.com) to introduce a sending delay between packets or simulate general operating conditions.
Again, depending on your test objectives you could also write a TCL script to fluctuate the QoS parameters over time.
HTH,
Colin
