Here's my short analysis of the current algo feel free to rebuke it if you have a better one.
'Pages' that is a dead end for a key term will rank high.
Example: Red widget
500 pages linking to siteA.com/page.html with the anchor red widget...
where siteA.com/page.html is further linking to siteB.com/page.html
siteB.com is the dead-end for anchor 'red widget'.
In the old algo siteA.com will normally rank higher because of the backlinks and perhaps all the in-page factor.
The current algo however will not rank siteA.com higher because it passes the term 'red widget' to siteB.com therefore siteB.com is the absolute authority about red widgets.
The beauty of this algo is that it will seek all pages that are dead-end for the specific term hence the highest relevency. Sites such as .edu, .org, .gov, and other purely informational/tutorial site.
Those on some very specific query you end up with very high relevance serp.
However, this is just good in theory. If the current algo gives so much weight to theming of anchors it neglect to consider the ff.;
1. The relevancy of the source of the backlink. It seems to be that the current algo is totally ignoring content or have weigh down the value of it. Backlinks from blogs, forums, guest books, and other sources that are not even remotely relevant to the term are given equal weight value to their anchor.
So if a general blog or a spammed guest book link out with the anchor 'red widget' the receiving page is deemed highly relevant to red widget assuming it doesn't out-link again the term red widget.
2. Internal backlinks are ignored or weighted down as well. Sites that have multiple pages related to 'red widget' are at the mercy of the external links. Even if the site itself is truly relevant to red widget but if the the backlinks doesn't reflect that then the site is deemed unrelevant to red widget. The situation is further aggravated by the normal navigational cross-linking. If you happen to out-link 'red widget', that page loses authority, like you have given that authority to some other page. Hence the phenomenom of previously high ranking pages dropping out or buried in the serp.
3. Portals, Links, and Directories are considered valuable resource for the specific terms...it points out the site that is higly relevant to the term. Most often portals/directories/links have high numbers of backlinks which are mostly copycat by other portals/directories/links. With the current algo P/L/D's wouldn't rank high because they out link the term.
Problem however, which the algo didn't take into consideration that most small P/L/D are 'not up to date or poorly maintained' and most often the out-links are dead itself. When this situation happen the said Portal, Link, or Directory becomes the absolute authority for that specific term because that is where ther term ends hence sometimes we would see directories/links/portals high up in the serp.
4. Lastly - a gift to spammers. Most spam sites are set-up where their back-links are term specific and the site itself doesn't link out. Since the theme of the source of those backlinks have no more or less weigh but their anchors are like gold the vote for this throw-away domain suddenly becomes very valuable.
Imagine if you have 1000 backlinks from spam source all using the same anchor. The throw away site will rocket to the top.
I've witness this already with the highly competitive terms where spam is rampant.