Making a strong password AND remembering it is like eating while talking. You choke. So the same thing might happen if you have a p455w0(R).|L1K3thys and someone cracks it. I'm just not sure if it's actually true. Are these leet passwords more crackable than completely random passwords that a random password generator makes? Are there any leet password crackers out there? Is there a way to safely simulate a penetration test on some offline leet passwords?
I think Trey Blalocks answer is great, but I would like to complement it with some math.
If your password is randomly picked from the 171,476 words in the Oxford English Dictionary you get log2(171476) or about 17.4 bits of entropy.
Lets assume there is about 4 natural leet substitution in the average word. Randomly either doing or not doing each substitution adds one bit, so adding the leet would increase the entropy by 4 bits, meaning it takes 16 times as long to crack. (If you just use leet for all available substitution you just add one bit - the password is either leet or not leet.)
On the other hand, a completely random 8 character alphanumeric (upper and lower case) password has log2(62^8) or about 47.6 bits of entropy. That means it takes a bit more than a billion times as long to crack!
So adding leetspeak is slightly better than just taking an english word, but it is not nearly as good as randomizing.
Cracking libraries do include common Leet substitution algorithms and there are Leet dictionaries which can be used by tools like Hydra. There are also tools to convert an entire dictionary of words to "Leet-speak"
More importantly hashes are available for the most common Leet passwords and Leet word variations so if someone is cracking a large password dump of these against a very large set of pre-hashed words which include Leet passwords in use they are very likely to find matches.
Finally a better way to determine real-world consequences might be to look at password dumps which have already occurred that also included Leet passwords. The proof of them being cracked is visible in a real world password dumps that have gone public. Likewise their presence in common hash tables (MD5 and SHA1) would also lend likelihood to them being cracked easily.
Obligatory and extremely relevant XKCD reference:-
What I'm inclined to do is combine both approaches but I don't think that makes a ton of difference. Another approach would be to utilise the first letter of every word in a favourite poem or song. NGGYUNGLYDNGRAADY etc...