Has anyone tryed these?
http://www.rflinx.com/products/filters/2400/diplexer/
Sorry, just ran into this but will reply because our results w/ the earlier model of diplexer sold by RFLinx are excellent, worth sharing. These are somewhat costly but highly effective in places where you may find yourself boxed-in due to physical constraints.
Specific case is that we had an poorly arranged legacy AP buried in the hinterlands, equipped w/a single Cisco 350 serving in AP/bridge mode, w/ a single omni antenna mounted high in the top of a mango tree. The local customers begged us to take over this device as otherwise they’d have no connectivity whatsoever (no landline phone, cellular, electricity, no nothing in that part of the world) . We’ve inherited several installations like this and normally we’ve waited for the antenna to be blown off the tree before doing any revisions. What with height and morphology these trees are not climbable by normal humans and tree climbers charge upwards of $1k to redo the mounts and reinstall antennas.
In this situation the antenna remained stubbornly attached to the tree and meanwhile service to customers utterly sucked. So we inserted a diplexer into the system, w/ a 112 serving backhaul duty to another stick in our system and the Cisco remaining in place due to yet more legacy issues having to do w/Cisco WGB radios.
Bottom line is that with both radios busy the noise floor on the 112 is at -105 (yes, because of extreme vegetation we’re running amps on both radios, apologies to noise sheriffs…) and we’re able to push plenty of traffic through the rig. Insertion loss is almost negligible; Cisco AP reports in “percent” signal and we saw a drop of about 2-3 Cisco mystery percentage points on clients. Meanwhile retry rates plunged since the poor Cisco was able to concentrate on pure access duty rather than store-forward.
It’s a shocking waste to use a 2.4 channel for backhaul so when the antenna does drop out of this tree we’ll redo the whole mess, but in the meantime the diplexer has transformed our customers’ User Experience™.
We’re shortly going to experiment w/ 5.8 diplexers w/ an eye toward converting some inner links in the system to dual nstreme where we don’t have windload capacity for more dishes. I’ll post back here when we have some results to share.
I’m surprised more people don’t take advantage of diplexers. They’re a staple item in the ham radio community for stuff like 2 meter band FM repeaters, which often run 50 or more watts into one coax line, shared with a receiver only 600 KHz away.
Here’s the unit dbostrom refers to. As he mentions, it would have been pointless to dump more resources into the top of a frigging tree. The diplexer is the silver block at the bottom.
Edit/update: To clarify, the power strip seen here is being fed from the relay host’s resedential PV system. When we say ‘no electricity’, we mean, of course, off grid. Since we put this in, were now seeing Skype traffic blooming down there. Pings back to our NOC (50+ mile circuit) from there, even when the clients are busy, run around 8-10 ms, with very little jitter.
I would love to have a diplexer that can do 5mhz or 10mhz bandwith channels in 2.4Ghz and some for the 5.x Ghz range too.
The off-the-shelf versions of these are tuned for standard 802.11x channels, so they’ll work just fine for narrower channels. 
For 5.x ghz check out http://www.lorch.com/wireless2.html
$540USD for a 5745/5810 combo. They’ll do custom-tuned variants as well.
I should also mention that RFLinx (an excellent outfit, with fully characterized and honest products) is also willing to mill these. The diplexer in the image is an RFlinx product.