Do we have another case of Schizophrenia here?
'Venkat" start the post and "Obalid" add explanations.
If you separate the I/O with 245s the input is not 'shared' any more. That said, you still have to be darn careful what you do in 'main' and what you do in interrupts.

One also wants to ensure that an apparent short of two of the port pins via one or more closed switches in the keypad matrix does not disturb the ability to communicate to the LCD module.
I have actually seen that one. If the user pressed two keys simultaneouly the whole thing went beserk.

There is so much that need be considered when making shared I/O that my current attitude (see reason in Michaels post) is that it is not worth even attempting shared I/O due to the probability of something like the above happening due to a slight oversight.

If you do not have enough I/O pins and more pins and plenty internal memory does not give you enough, use a CPLD or - as I am currently doing - throw a couple of $1 LPC chips in and talk to them over IIC. In addition to more pins that has the added advantage of giving you more timers, more interrupts etc.

It takes a lot to get to where you need extra I/O e.g. the Cygnal 126 has 8k RAM and 128k Flash inside which leaves 8 ports free.

Erik