Distance is also quoted as an issue. 1-1.5km being stated as a limiting factor due to the round trip time (speed of light) over these distances. This is not what people have experienced and PeerTech explains why. There are many examples of networks working well over these distances and not experiencing problems with retransmissons.
The 802.11 RTS/CTS protocol is the official solution to this. It doesn't work. In fact, theory (back up by lab testing) shows that it makes things worse.
External arbitration is another possibility. Karlnet's TurboCell and Proxim's WORP on their Tsunami product range, being two of the commercial solutions. Frottle and WiCPP being two open source projects to solve this.
Multipoint bridging is offered in many cheap 802.11g basestations, but the solutions are all proprietory and many only work with products of the exact same type, not even with other products from the same manufacturer. A node failure in a years time could lead to the need to replace all equipment or to build a special case connection for that node because that product has gone end of life.