vorbis encoder performances under arm platforms

Hi, 

i am trying to do vorbis encoding on an arm-powered device (an iPhone), but i am getting really poor performances. 
I am using vbr encoding and setup it with:
vorbis_encode_init_vbr(&vrb_info, 2, 44100, .2);

the cpu load seems not to change according to the quality, on a iPhone 3gS seems to average at about 60-75 % of the cpu time (i capture audio from theinternal mic and encode it on-the-fly to vorbis). 
Some questions i seem not to find a definitive answer:
- floating point operations are the main cause of poor performances?
- i see there's the tremor project for decoding. Is something similar planned for encoding (is it feasible?)?
- any compile-time optimizations i could try? (i am quite sure there is nothing i could do on this side)
- which is the fastest encoding mode using vorbis? i am using that vbr without bitrate management at the moment, is there something faster?

sorry to bother you,

marco