1. Writing C extensions is a horrible experience. Also, they have to manually yield priority so if they crash, the whole BEAM crash so the whole "fault tolerance" propaganda becomes a lie.
2. Erlang doesn't cooperate well with it's environment. Starting, supervising and communicating with external processes correctly is tough in any language but harder in Erlang.
3. BEAM:s memory usage is hard to predict, so running multiple vm:s on the same system can easily run you out of memory.
4. It's hard to run Erlang code as standalone scripts. Points 2-4 address the same issue which is that basically that Erlang's environment is a walled garden.
5. While Erlang is open source, it's mostly run by a group of developers at Ericsson and doesn't have much of a community. For example, it doesn't have a public bug tracker (maybe that have changed).
6. It's syntax is, objectively, butt-ugly.
7. Maybe most important: It is not a good fit for about 99% of all programming problems which are mostly sequential and doesn't benefit at all from rock-solid (Erlang) process communication.