Hmm, this made me look deeper into it again, and I found some clues which I never had before:
I have the habit to run apps which access sound device under aoss wrapper to prevent them from being blocked by another app accessing sound device.
Running under aoss wrapper:
32228 31277 tc R 3928 0.7 0 29.2 mpg123 /mnt/sda1/sound/rumble.mp3
32191 20893 tc S 21644 4.1 0 7.5 mplayer /mnt/sda1/sound/rumble.mp3
Running each one exclusively without aoss wrapper indeed there seems to be no significant difference:
2207 31277 tc R 2400 0.4 0 5.2 mpg123 /mnt/sda1/sda3/sound/rumble.mp3
2310 20893 tc S 20404 3.9 0 4.9 mplayer /mnt/sda1/sda3/sound/rumble.mp3
The plot thickens when running mpg123 with "-o alsa":
2542 31277 tc R 3856 0.7 0 13.7 mpg123 -o alsa /mnt/sda1/sda3/sound/rumble.mp3
using much more cpu than without "-o alsa", but much less than under aoss wrapper.
When running mpg123 with "-o oss" then it stays in same range as without any parameter:
4563 31277 tc R 2400 0.4 0 5.2 mpg123 -o oss /mnt/sda1/sda3/sound/rumble.mp3
I am a bit puzzled now, does that mean that without parameter mpg123 would use alsa module oss emulation and use much less cpu than with native alsa??