Just for giggles, try increasing the number of $55s you send to..say 8. Next pick a more balanced header character like $AA. You can also put a small character delay using DEFINE (Maybe 1 mS). If this helps, you need more of a preamble to balance the data slicer.
What are you using for antennas? Usually the companies range test their units with twelve element Yagis. Well, maybe not that bad, but the test data is usually taken in an optimum noise free environment with no multipath. These parameters don't exist on our earth, and you should probably halve most manufacturers range estimates. I've had to test most offerings in the OOK market, and upgraded to FSK for my more serious work, and better impulse noise immunity.

Ron