People often claim that meat is inherently inefficient because we feed animals grain and it would be better to eat the grain ourselves:

Beef cattle, weighing in at around 900 pounds as an adult, require around 80 pounds of food each day, 18 pounds of which is grain (agric.gov.ab.ca), (most beef cattle in the U.S. are grain-fed (wikipedia)). Beef cattle live for only three to six years before being slaughtered (wikipedia). Calculating only a three-year lifespan, that means the cow would have consumed 18 pounds of grain per day x 365 days per year x 3 years = 19,710 pound of grain during its life. But, of course, the animal is not born full grown, so we will cut this number in half to 9,855 pounds of grain consumed in its lifetime. Typically 62% of the weight of the animal ends up as meat (dead link) [1]. So for our 900 pound example, we would have around 558 pounds of meat. 9,855 pounds of grain divided by 558 pounds of meat is 17.6 pounds of grain for each pound of meat. -- source

There are some problems with this approach, but first let's redo the numbers. This is based on cattle having the same feed continuously, but the standard way is to feed them on forage (hay/straw) when they're smaller and switch them to grain for a shorter time to quickly fatten them up. Because they need food proportional to their current size, linear growth as described above would be much less efficient. Reading UMass Extension, they generally stay on forage until they weigh ~750 pounds, and then they go to a feedlot where they're fed large amounts of grain to quickly get them to ~1300 pounds. That's ~550 pounds of weight gain, and they say that takes ~3000 pounds of corn. Using the same "62% meat" number from above this gives us a conversion of 3000 pounds of corn to 800 pounds of meat, or 3.7 pounds in per pound out.

This is strange; isn't "90% of energy wasted going up a trophic level"? How can we get numbers better than 10 into 1? Several things are happening. First, the 90% number people give is an average over the world ecosystem, and includes a huge number of plants operating at 0.1%-2% efficency, so it's possible for corn into cows to be much better than the mean. Second, the trophic level claim is about food energy, but we've been using pounds. Beef is about 1000 calories per pound while corn is closer to 600, however, which moves the numbers in the opposite direction: 2.2 calories of corn in per calorie of beef out. And third, we haven't included the calories the cattle get from the forage they're eating before they're moved to the feedlot. This makes sense, because people can't eat it and you can grow it on marginal land that wouldn't support corn, but it also means that claims about efficiency based just on grain are missing something important.

Another question is whether the corn we feed to cattle really is something we could just eat instead. While feed corn is cheaper than regular corn it's not much cheaper, not enouch to matter here. But not all the 'corn' cattle get is raw; a lot of it is waste from other industries. In making alcohol or high-fructose corn syrup there's partially used corn left over that's fed to cattle. In some places this is sigificant, but as far as I can tell it's a low proportion overall.

So let's say I become a vegan or vegetarian and substitute seitan for my beef. Seitan is made by 'washing' flour until all you have left is the gluten. While flour is 10-15% gluten, seitan uses high-gluten flour, and so about 14% of the original wheat becomes seitan. This is about 7 calories or pounds of wheat per calorie or pound of seitan. This is worse than the 2.2 calories or 3.7 pounds of corn per calorie or pound of beef we got above.

Update 2013-11-30: This misses that beef and pure seitan have different protein densities. Updated math.

If you want to eat a diet that's maximally efficient in terms of energy transfer all the way from light up to calories in your food, that is going to be a vegetarian diet, but not all vegetarian diets are efficient in this way.

(There are lots of other reasons people give for not eating meat and this is very much not an attempt to be exhaustive. This is a look at a single widely-quoted statistic, "Xlbs of grain per lb of meat".)

[1] While 62% is a reasonable number for the fraction of the cow that's turned into meat, the remaining 38% isn't waste. We use almost all of it in various forms in other industries.