I noticed that album RG values are sometimes higher than RG values for some of the tracks on that album. For example, 2nd track has RG value of -2dB, but RG value of the album on which that track is = -1dB. Shouldn't album RG value be equal or even lower than the RG value of the loudest track on the album?

No, album RG is a kind of "average" value, like track RG is a kind of average value, not referred only to the loudest point of the track.
It's not normalization of the track or of the album...

(this is a simple and innaccurate post, but I need some coffee )

edit: spelling

Please read the section "3. Statistical Processing" on this page:
http://replaygain.hydrogenaudio.org/calculating_rg.html

Remember, for the album gain the statistical processing is done over the entire album, so this "value 5% down the list" can be quite different from the one for a single track (e.g. 5% of 60 minutes of audio versus 5% of 3 minutes of audio).

Fair enough, but if for example most of the tracks on the album are very quiet, and few is VERY loud, then (if I understand things correctly) album RG value can get positive and lead to clipping on that loud track (even if originally it wasn't clipped).

Yes positive replaygain values can lead to clipping. That is why some programs also calculate the peak value. So it can make sure that the peak doesn't go over the maximum allowed value.

Foobar2000 does this if you select "use peak info to scale down tracks that still clip after applying replaygain" under playback.

Is it part of the RG standard or just an extra feature of some programs? If it's just an extra feature, then IMO it should be implemented in RG standard (I assume that 2nd reason for RG existance is to protect from clipping).

It's not part of the outdated standard, which only mentions AlbumGain, TrackGain (==Audiophile, Radio) and track-peak.
But all implementations I know of also support album-peak values.

The official replaygain documentation is largely outdated and should be updated to at least reflect real-world-usage at some point. 2BDecided (original Replaygain developer) started a thread discussing further development of the ReplayGain standard and the possibility of a ReplayGain2 a while ago.