The in the summary described behavior happed on a production server, but we could reproduce ist on our development server, too.

I logged in to the secondary server:
SECONDARY> rs.slaveOk()
SECONDARY> show dbs
blablub 0.203125GB
hm 0.203125GB
hm_history 0.953125GB
hm_idealo 0.203125GB
hm_logs 2.2021484375GB
hm_solr 0.203125GB
hm_tracking 9.94921875GB
hmdev 0.203125GB
local 10.1982421875GB
sequences 0.203125GB
test 0.203125GB
tests 0.203125GB

Switched to hm_history database:
SECONDARY> use hm_history

Then started the compact command on two existing collections;
SECONDARY> db.pricehistory_2010.runCommand("compact")

RECOVERING>
SECONDARY> db.pricehistory_2011.runCommand("compact")

Until this step everyting runs fine, but after running a command on a collection wich does not exists the server remains in state recovering. Our production server remained in this state for 5 hours, so that we needed to restart the service.

RECOVERING> db.dffsdfsf23422eqw.runCommand("compact")
{
"assertion" : "namespace hm_history.pricehisto does not exist",
"assertionCode" : 13660,
"errmsg" : "db assertion failure",
"ok" : 0
}
RECOVERING>
RECOVERING>
RECOVERING>
RECOVERING>

Output from rs.Status:
RECOVERING> rs.status()
{
"set" : "hmdev",
"date" : ISODate("2012-03-23T13:47:03Z"),
"myState" : 3,
"syncingTo" : "devdb.hitmeister.intern:27017",
"members" : [
{
"_id" : 0,
"name" : "devdb.hitmeister.intern:27017",
"health" : 1,
"state" : 1,
"stateStr" : "PRIMARY",
"uptime" : 2238,
"optime" :

,
"optimeDate" : ISODate("2012-03-23T13:10:10Z"),
"lastHeartbeat" : ISODate("2012-03-23T13:47:02Z"),
"pingMs" : 0
},
{
"_id" : 1,
"name" : "devdb.hitmeister.intern:27018",
"health" : 1,
"state" : 3,
"stateStr" : "RECOVERING",
"optime" :

,
"optimeDate" : ISODate("2012-03-23T13:10:10Z"),
"self" : true
}
],
"ok" : 1
}