Here's an example of Google doing better then a discovery system. First and foremost your mileage may vary. This is a very specific example but endemic of the landscape we find ourselves in.

The Setup

Since I have some tiny amount of domain knowledge now this doesn't come from an abstract Librarian assessment of what results are best. Last summer I read some work on an aspect of a mathematical process named after Pollard written by a someone named Teske. So with a simple keyword search:

teske pollard

I tried to get back to those readings using two different tools

Google Results

Screen cap of results

1. The paper I wanted. Full text, in postscript to boot. Hosted at the affiliated school's departmental website. (Author's hosted copy)
2. Second great result. Full text and hosted at affiliated schools deparmental website (Author's hosted copy)
3. Abstract (in PS format) of conference paper described in result 4. Once again hosted at school's website
4. Conference paper published two years after what I was looking for. Probably a combination of paper 1 and 2
5. A paper citing the paper 2 and 4

Discovery Results

Screen cap of results

1. The paper! Yeah, but hosted through JSTOR, ie a link I couldn't share with anyone who wasn't affiliated with a school that has a subscription.
2. A paper on a similar topic but not what I was hoping for about 2/3 relevant
3. Perhaps some research in a similar topic but not what I was hoping for, about 1/3 relevant
4. Just the citation (not full-text) of result #2 from Google. A novice searcher would presumably give up at this point as the OpenURL/ native full-text didn't lead anywhere
5. Not close enough for me to click on.

Verdict

Putting this together I was stewing in my mind over what sort of scathing indictment I could levy again the discovery system... It was out of touch with reality, didn't understand the topic, missed the point of what I was trying to find.  Ultimately though I think discovered what I was.  The discovery system is constructed / predicated on respecting the formal publishing lifecycle such as it is. Things don't count here unless it is birthed through the process. Case in point Google Result #1 and #2 are not the formal copy of a finished product but it is what I wanted. Extrapolate a bit, if we as a profession are pushing OA via Repositories then our discovery systems should reflect that in some respect though, right? The utility of pre-press/author copy will be different for different disciplines but full-text is full-text and when it comes to it I'm happy that I have the PS files I was gunning for with 13 keystrokes.

Over exaggerating?

Oh you bet. However anytime this happens that is one more tiny bit of preference felt towards Google. Happens enough times and it leaves a lasting impression. Our edge/utility/je ne sais quoi/differential errodes and people go to the G first. In my year-ish doing a Master's degree this scenario has come up again and again.