I think the problem is that boost introduced a version of lexical_cast for 
whole arrays and therefore boost::lexical_cast<T, std::string> is ambigious?

The following patch resolves the compilation issue but I didn't test yet if it 
breaks anything else. However it would be nice to find a way to still use 
std::transform / anything more elegant which is backward compatible.

--- a/include/votca/tools/tokenizer.h   Sun Dec 02 21:10:09 2012 -0700
+++ b/include/votca/tools/tokenizer.h   Sun Jan 27 03:09:10 2013 +0000
@@ -93,8 +93,10 @@
         std::vector<std::string> tmp;
         ToVector(tmp);
         v.resize(tmp.size());
-        transform(tmp.begin(), tmp.end(), v.begin(),
-            boost::lexical_cast<T, std::string>);
+        typename std::vector<T>::iterator viter = v.begin();
+        typename std::vector<std::string>::iterator iter;
+        for(iter = tmp.begin(); iter!=tmp.end(); ++iter, ++viter)
+            *viter = boost::lexical_cast<T, std::string>(*iter);
     }

 private:

This issue was closed by revision f26d7ee778d6.