Updated cgminer and tested it successfully mining on the only stratum ltc pool:
https://bitcointalk.org/index.php?topic=28402.msg1419834#msg1419834
(lots of other new scrypt goodies in this release too).

New release: 2.10.3 - 26th December 2012

What a difference xmas makes.


Human readable changelog:

When switching from a stratum pool to another one, the last few shares were dropped and the share submission failure message was displayed.
There may have been stratum messages waiting to be parsed and cgminer would only parse them once a new message was received. This fix could potentially give even lower rejects!
There is now a [Z]ero stats option under the display menu which resets all the visible statistics.
Fixed the best_share displayed bug.
The current pool target diff is now shown in the status line like so:

The current block target diff is now shown in the status line like so:

Block solve detection is now supported with scrypt mining as well (BLOCK! written at end of share and solved blocks listed under pool stats).
Fixed a crash with rolltime pools.
Stratum support for scrypt mining.


Full changelog:

- Do not give the share submission failure message on planned stratum
disconnects.
- Parse anything in the stratum socket if it's full without waiting. Empty the
socket even if a connection is not needed in case there are share returns.
- Provide a mechanism to zero all the statistics from the menu.
- Display the current pool diff in the status line.
- Display block diff in status line.
- Generalise the code for solving a block to enable block solve detection with
scrypt mining.
- Generate the output hash for scrypt as well and use the one function to set
share_diff.
- Use the flip80 function in regeneratehash and the correct sized hash array.
- Use one size for scratchbuf as a macro in scrypt.c
- Stage work outside of the stgd lock to prevent attempted recursive locking in
clone_available.
- share_diff needs to be performed on a BE version of the output hash to work,
leading to false best_share values as spotted by luke-Jr.
- Remove the unused sha224 functions.
- Use the flip functions in hashtest.
- Simplify the setting of the nonce data field in work on submitting nonces.
- Scrypt code does not enter the hashtest function.
- Go back to cloning available work under staged lock.
- Updated links to AMD APP SDK
- Updated link to ADL SDK
- scrypt_diff uses a uint64_t as well.
- Correct target for stratum support with scrypt mining.
- libztex: fixed a typo
- libztex: check returnvalue of libusb_claim_interface() and release the
interface in case of early exit

Should be fixed in git now.

Alas the best share bug is NOT fixed, sorry.

Meh the lock got lost in the rework for 2.10, will fix in git. This is likely your problem after all, thanks!

I think there's a valid point here though. His version of your code is not the issue here but rather a problem with the hash when it hits the share_diff function in cgminer.

For the first function, you should note
So adding another lock does nothing and only risks recursive lock deadlocks or livelocks.

For the second value, work_block is simply an integer value and even if it's wrong can not lead to a crash since there's nothing being dereferenced at any stage.

I pushed variance to the maximum by setting my worker diff to 10000 which would only average 5 shares per day at my 2.7GH and got lucky with this:
 [2012-12-23 15:48:21] Accepted 00001e9b Diff 548K/10000 GPU 0 pool 0

POT Value    0.46005046 BTC
PPS Percent    666.00%

I only would earn .35BTC per day mining PPS, and gotta love that satanic percentage 

I haven't posted a release with the fixes yet so not sure what you're using. Only one part of the fix has even made it into git. Other changes need to follow.

From the [p]ools menu, if you select one pool, it will show up if a block was found for that pool, otherwise it will say nothing about blocks if you found none.

Increasing the work difficulty to a fairly high level makes this much more like mining solo but at much lower difficulties Trying to decide how high you should set it and accept more variance is the trick though

Yes that makes sense thanks. I shall fix it.

Now you're smoking the ganja.

Care to explain why cgminer natively doesn't work? Is the problem in cgminer or the pool's stratum implementation?

W00t that one share was worth more than 86,000 PPS shares 

That's just the nature of how long the timeout needs to be on a raw socket to reasonably know it's dead with stratum. There is nothing you can manually change to alter that apart from not putting a dead pool first in line. This is not a new issue with 2.10.2, it's just that btcg is suffering a DDoS. If it were a getwork pool it could take up to 60 seconds. With stratum it's over 2 mins.

That change is in the latest cgminer, 2.10.2

Does it actually fix anything?