OK - a number of points. One - a lot of grain is still grown in unirrigated land. Secondly, though we irrigate more than we should that does not mean that irrigation has to be cut down to zero. The point that some areas are irrigated with unsustainable quantities of water does not translate into "we need to replace crops grown there with grasses". With more efficient irrigation methods (highly efficient sprinklers, drip irrigation) we can obtain the same or better results with much less water. And then if engage in soil building we can cut water use further. And then we can capture run-off, purify it with reverse osmosis, and reuse that water. In short we can irrigate, but via withdrawals of small enough amounts of water to be permanently stable. Most of the land under cultivation could continue to be under cultivation, but cultivated better. In the great plains, where grain and soybeans are currently grown in short rotation we could use longer rotations - legume, grain, fiber crop, green manure. Irrigate them efficiently and sustainably. Arizona lettuces (and Arizona cotton) - well there you have a point. It does not make sense to grow water intensive crops in Arizona. In terms of fruits and vegetables a lot could be grown within cities and suburbs. At any rate it is grains and legumes that take up most of the row crop acreage. Lettuces, oranges, non-starchy veggies, and fruit - generally these take up a tiny portion of the land. U.S. population is expected to peak at 400 million. So basically you could feed the expected U.S. peak with about half the acreage currently used to grow grains and pulses used to feed animals. Plus you have the land that currently grows grains and pulses for direct human consumption. Along with existing grazing land, maybe the other half should be put into grasses for something along LPs proposal. Bison or whatever is appropriate.
The numbers may be different in other parts of the world. But most areas of the world where grain and pulses are grown today and continue to produce either grain and pulses or some other form of vegtable protein and carbohydrates in ways that are sustainable in terms of soil building, water use and biodiversity. So we can sustainably produce vegetarian protein for a population of 9.5 billion. I'd be interested to take LPs numbers on bison production, and look at converting all the land currently used to produce food for animals worldwide into grasslands, and directly grazing appropriate creatures (bison, goats, sheep, emus, whatever) on it. I wonder if that could provide protein for 9.5 billion. Also I wonder about the greenhouse effects. Even green grazed animals produce a worse greenhouse effect than low-input vegetable protein. Even when green grazed animals build rather than erode soil, there is there contribution to the nitrogen cycle, plus methane released in burps and farts. Most existing analysis considers the the methane but not the nitrogen.
David Pimentel is where I got the 800 million figure from http://www.news.cornell.edu/releases/aug97/livestock.hrs.html. Note Pimentel looks at grain only, but obviously a grain/pulse mixture produce a complete protein and also (by using legumes) avoid the need for nitrogen fertilizers (along with green manure of course).