These chips in particular, and apparently others that use similar output stages, are exceptionally sensitive to impedance matching on the output. With a moderate mismatch (1.5-1) you may see a third or fifth harmonic that exceeds the carrier level (!), as well as a significantly depressed carrier level.
As it happens, there exists a transformer specifically designed to act as a filter/balun for the Atmel parts. I can't swear to it for other chips. Getting this little piece of information, and redesigning our antenna for a closer match resulted in a 10dB boost in radiated power along with the harmonics being below -45dBc instead of +15dBc.
Two resources that you should know about if you are in this or a similar hole, especially if you are within shouting distance of Sparkfun:
1: http://www.emitestlab.com/ Dennis King is SUPERB, and his pricing per half day is surprisingly reasonable. He uses a GTEM so scan time is fast and the results correlate well with our OATS.
2: http://www.ccn-i.com/Welcome.html If Don DeGroot can't figure it out, you are in one deep and UGLY hole.
Both are in Longmont, and I've been working with them for over a year now.
No direct relationship, just a very satisfied customer.