This came up today in a conversation with a postdoc in my lab, Graham Reynolds.
There are probably more elegant ways to do this, but one simple solution for plotting error bars in two dimensions is provided by just calling the gplots function plotCI twice. Here's an example:
# simulate data
# here is a completely arbitrary simulation model
# for the standard error/deviation
# first get the limits for x & y +- error
# now plot twice, remember to set add=TRUE
The result is not too bad: