I suspect it boils down to two things: the number of ways in which things can compete and the perceived level of normal consumption.
Starting with the latter: people who drink coffee mostly see it as near-daily consumption. If they're thinking about it at all, they're not thinking about "will I have coffee", but rather "which coffee to have". On the other hand, many software buyers have "don't buy anything" as a very real possibility in their minds. If I "have to buy it anyway", the difference between $2 and $4 per day doesn't seem important because people are surprisingly bad at multiplying by 365.
Coffee also presents a much less complex competitive landscape than software (which is not to say it is less competitive; I strongly suspect the reverse is true). As far as I can tell coffee shops can compete on price, location, ambience, brand, and flavor. My guess is that, above a fairly low bar for flavor, location is the dominant competitive attribute. "Am I going to walk an extra block to save $1?" Again, people don't know how to multiply by 365. I suspect coffee shops strategize thusly: "I'll get whatever customer base is closer to me than to competitors, and fighting on price is pointless. I might get some marginal purchases by improving branding and ambience, thus persuading people to stay longer and buy more." So everyone pays $4 at the nearest half-decent coffee shop and all is peaceful in the universe.
Software, on the other hand, is an incredibly complex offering and I think that complexity leads to a perception that diligent consumer research will be rewarded with enhanced satisfaction or reduced expenditure. The problem here is that the reduced expenditure is easy to quantify and guaranteed prior to purchase, whereas the enhanced satisfaction is poorly quantified and feels uncertain unless you're dealing with a market-leading strongly-branded product like Angry Birds. So, unless your'e buying Angry Birds, you'll spend some time looking to see what other apps are cheaper with "the same" features.
Hm. I feel like even after three rounds of editing I've achieved a very low insight-to-sentences ratio. Sorry :-)