I've always found it far easier to scan through some psuedo-code and read a few paragraphs explaining it. At this point, I can usually look at the loops, recursive calls, and "structure", and get a pretty good idea of how it's working, what it's big-O run time would be, and why. Full blown formal proofs and theorems just don't help my understanding.
I agree the math is important for research and writing papers, and proving one algorithm is better than another, but for practical purposes (i.e. actually implementing an algorithm in real code) it's just not very helpful. And the reality is that most people learning about algorithms are going to spend a lot more time implementing algorithms than they will writing papers and doing algorithm research, so I'm not sure a math heavy book like CLRS is really the best undergraduate text.
On the other end, CLRS is far from perfect. My two issues with CLRS are:
1. Many critical topics are missing entirely while lot of non-critical stuff have gotten in to. There is a discussion started out by Professor Corman at Quora for suggestions for 4th edition here: http://www.quora.com/As-we-start-planning-the-next-edition-o...
2. CLRS doesn't often try to make you understand the topic or algorithm. For example, for BST it will just say this is how you find successor node and it's entirely your job to think why this is the best way or why other alternatives won't be good. The way I deal with this is stop reading after chapter introduction and think about for a while how I would have solved problem and then read explanation in the book. This often illuminates subtle points of algorithms. I ended up annotating huge swath of text in my own PDF copy.