I've learned a bit more about some of the features of the HP 8590L spectrum analyzer I got off eBay a while ago. One feature it has is the ability to calculate channel average power and the adjacent channel power ratio (that is, how many dB down the power is on the 6 MHz above and below the channel of interest). And at the preferred settings of the exciter, the output power is about 15 watts, with the adjacent channels being down about 35 dB. That's a little more out-of-channel garbage than I had hoped for - a bit more than 20 mW of ERP. But even so, on 900 MHz that small amount of power should die easily within half a mile of the site. Heck, I've transmitted with the bare exciter at approximately that power level and been unable to receive it half a block away from home. So I'm not going to worry about it. As always, reducing power will reduce the reconstituted sidebands, though the bare exciter has sidebands that are only about 48 dB down from the main signal, so there's only so much more improvement to be had.
Another bit of good news is that I spent some time "boiling the dummy load" this evening. Without the muffin fan, the heat sink of the amp gets so hot that you can't comfortably touch it. But with the muffin fan sitting on top and running, the heat sink stays almost as cool as the ambient air temperature (almost). So that bodes really well for the health of the amp when it's put in service.