Bug 1007504

Somehow I think, this is more or less expected on normal workstation hardware, as it just takes time for disk IO. Esp. horizon can't do anything in preventing this wanted IO.

this is not a disk I/O issue. 
seems like this would be an issue when we move to testing loaded environments. 
Please move to next version if you don't think it can be fixed in current version but it's most surely a bug

Horizon just responds to underlying services. If they don't answer (for whatever reason), horizon waits (until timeout)

When reproducing this, my machine gets a significant load avg (30), and esp. high iowaits (85%). 

you can try that yourself (even in shell):
#!/bin/sh
function create() {
start_create=`date +%s`
for i in `seq 1 8`; do cinder create --display-name myvol$i 1 ; done
end_create=`date +%s`
echo $end_create - $start_create | bc
}


function delete() {
start_create=`date +%s`
volumes=`cinder list | grep "myvol" | cut -d '|' -f2-2 `
for i in $volumes ; do cinder delete $i; done
end_create=`date +%s`
echo $end_create - $start_create | bc
}

create
delete


And wait 10 mins, until your computer is usable again ;-)

it seems, cinder delete using a loopback device etc. is a very expensive function.