ejabberd burning cpu and eating memory

... and already tried with max_fsm_queue.
I've pymsn-t and pyicq-t running for years now, yesterday I installed spectrum transports (currently for testing only), and ejabberd crashed overnight.
Right now the situation is like that: when no spectrum transport is running, ejabberd uses almost 0% cpu and about 613M Ram. After starting a spectrum transport (only one, doesn't matter which one) the cpu load of ejabberd increases to 100% after a few minutes.
I played around with the configuration options in ejabberd, tried almost everything - reads like this now:

 {5351, ejabberd_service, [{access,local},{max_fsm_queue, 5},
                           {max_stanza_size, 65536},{shaper, c2s_shaper},
                           {backlog, 2},{service_check_from, false},
                           {host, "msntesting.domain.tld",
                           [{password, "random"}]}]},

I changed my ejabberd loglevel from 2 to debug and there are very much messages like:

=INFO REPORT==== 2010-04-18 14:15:37 ===
D(<0.10861.0>:ejabberd_router:301) : route
        from {jid,[],"msntesting.server.tld",[],[],
                  "msntesting.server.tld",[]}
        to {jid,[],"msntesting.server.tld",[],[],
                "msntesting.server.tld",[]}
        packet {xmlelement,"iq",
                   [{"type","error"},
                    {"to","msntesting.server.tld"},
                    {"from","msntesting.server.tld"},
                    {"id","uid:4bcaf79d:1ded0839"}],
                   [{xmlelement,"ping",[{"xmlns","urn:xmpp:ping"}],[]},
                    {xmlelement,"error",
                        [{"code","405"},{"type","cancel"}],
                        [{xmlelement,"not-allowed",
                             [{"xmlns","urn:ietf:params:xml:ns:xmpp-stanzas"}],
                             []}]},
                    {xmlelement,"error",
                        [{"code","405"},{"type","cancel"}],
                        [{xmlelement,"not-allowed",
                             [{"xmlns","urn:ietf:params:xml:ns:xmpp-stanzas"}],

Any ideas whats going on here/how to fix this?

Edit to add some additional information: I'm using Ejabberd 2.1.3 compiled with Erlang R13B04 on Gentoo. I did set max_fsm_queue to values between 5 and 1000, always the same result. Also I tried to disable mod_ping for the domain Spectrum is connecting to.
If I should post more of the logfiles or try something else (perhaps without downgrading Erlang) just tell me, I'll do ;)