I think the "Erlang sucks for strings" argument is overblown, and it's rarely if ever backed by real world experience and/or data. Erlang is fine for most string handling tasks you'll find in a typical (web) application.
We have to be honest identifying our tool's weaknesses, Yariv, even if we like them. EEP-009 (http://www.erlang.org/eeps/eep-0009.html) and EEP-010 (http://www.erlang.org/eeps/eep-0010.html) are extremely important for the continued growth and adoption of Erlang.
Erlang may not be optimal for applications that do heavy string processing, but the significance of that statement is overblown. Take typical webapps for example. What kind of string processing do they do? Very light stuff. They spend of their cycles slurping binary data from the database and sending it down a socket. Erlang is excellent for this usage. They occasionally take form inputs from a user, in which case they do some simple processing to ensure the input is properly escaped. Erlang handles that just fine.
Most of the "Erlang sucks for strings" arguments remind me of the "Ruby/Java is slow" arguments. They're sort-of true, but they just don't matter for most applications.
Please don't confuse this sentiment with any anti-erlangism or any assertion that Erlang isn't fantastic. I just want to be realistic, and I'd love to have faster string processing to make faster routing code for Fuzed (http://github.com/KirinDave/fuzed/tree/master).
Erlang has weaknesses, but certainly those are not it. Basically, Erlang excells at writing servers. Take a look at YAWS (http://yaws.hyber.org/) for a web server written in Erlang that can actually outperform more traditional web servers in very realistic scenarios.
If I had to name a weakness or two of erlang offhand:
1. Integrating foreign code with Erlang is very different from traditional FFI and requires some initial legwork (libraries like Erlectricity http://github.com/KirinDave/erlectricity/tree/master , for example).
2. String manipulation. Erlang doesn't have a real string type. Its string libraries and core string type assume using linked lists for strings, which is spectacularly inefficient. However, there is a movement and a proposal to change the string type to Erlang binaries and provide an alternative string library. Hopefully that will address the problem.
Other than that, it's a reasonably fast and rock-solid implementation of an FP language with a mountain of solid IPC library code integrated at the language level.