Have the same problem. I guess updating the same record again doesn’t help anymore. I tried inserting new records and it works, but not the update. I guess this is what is preventing:
When a record is created, or when a record is edited and did not previously meet the rule criteria
Correct me if i am wrong. I tried it and is not firing the rule the second time, after it is updated in the trigger. Wondering how to fool the rule to think that it didn’t meet the criteria previously. Guess thats not possible!

I hit the same problem and adjusted my code to create a new BatchObject (which I call Scheduled_Job__c) record every time it gets set off, instead of updating the same record. It gives me a log of what’s happened, and I then have a method, also run by the batch routine, that deletes old Scheduled_Job__c records to keep things clean.

This approach also works for me because I wrote my routine to process records in chunks that I know won’t set off the governor limits (always a battle), so the code generates a new Scheduled_Job__c to run immediately until all records are processed and then schedules itself again for the next night at midnight. You just need some way to track which records have been processed already, which I do by just adding a timestamp field to the records that I’m processing that gets updated when the batch is run. I’ve used it to run some typical batch processes and also to send out a dynamic email each night to a bunch of users, where I’m limited to sending to 10 email addresses at one time and about 40 people need to receive it.

It’s a hack on a hack on a hack, but it works! Luckily, I’m nowhere near hitting the limit on workflow time triggers per day. This is the trigger I’m using-

trigger InsertUpdateScheduledJobAfter on Scheduled_Job__c (after insert, after update) {
Integer intReturn;
Integer intMagicNum;

for (Scheduled_Job__c sj : Trigger.new){
if (sj.Run_Job__c == true){

//Create Site Log records for 7 business days in the future, if needed
if (sj.Method_To_Execute__c == ‘CreateSiteLogs’){
intReturn = clsScheduledJobs.CreateSiteLogs(null);
//This job is run in batches of 2 site invoices (the max that can be processed
//before hitting the governor limit), so less than 2 means it’s done
intMagicNum = 2;
}

//Send out Job Loss Report
if (sj.Method_To_Execute__c == ‘JobLossReport’){
intReturn = clsScheduledJobs.JobLossReport();
//Single email limited to 10 at a time
intMagicNum = 10;
}

if (sj.Method_To_Execute__c == ‘ScheduledJobCleanup’){
intReturn = clsScheduledJobs.ScheduledJobCleanup();
//This job always returns 0, shouldn’t need to be re-run anytime soon
intMagicNum = 1;
}
Scheduled_Job__c new_sj = new Scheduled_Job__c(
Method_To_Execute__c = sj.Method_To_Execute__c,
Name = sj.Name,
Run_Job__c = false,
Schedule__c = true);

//If the number returned by the method is less than the Magic Number,
//create a Scheduled Job record to run it again tomorrow night
if (intReturn < intMagicNum){

//Reset Scheduled Job record to midnight of tomorrow
new_sj.Next_Run_Datetime__c = Datetime.newInstance(System.today().year(), System.today().month(), System.today().day()).addDays(1);

//Otherwise, there are more records to process, so create a record
//to run it again immediately
} else {
new_sj.Next_Run_Datetime__c = System.now();

}

insert new_sj;
}

}
}

The general flow goes like this (I’ve simplified the debug output for brevity sake) …

BatchObject is modified, RunBatchJob=true.

Beginning triggerX on BatchObject (After Update)
Batch job runs and sets RunBatchJob=false

Beginning triggerX on BatchObject (After Update)
This time RunBatchJob=false, so no action taken.
Ending triggerX

Beginning Workflow Evaluation
RunBatchJob=false so criteria evaluates to true and time-based action is queued.
Ending workflow evaluation

Ending triggerX

Beginning Workflow Evaluation
RunBatchJob=true (!! apparently the original version of BatchObject) so criteria evaluates to false and time-based action is removed from queue
Ending workflow evaluation

How do I stop the Workflow Evaluation at the end from occuring and undoing my work? I’ve tried setting separate flags for “run apex” and “run scheduler”, and I’ve run the trigger before and after update, but before the trigger/workflow cycle can complete, my trigger has to update a field used in the workflow rule and this problem re-appears.

For the latest numbers, please refer to the FAQ, which i will be posting soon on Time-based workflow feature detail page on successforce.com website.

For example: if you want a batch job to fire every 2 hours, you could create 12 Batch Script objects, each with a Test Script Run Date 2 hours later than the last. The practical limit would be around 10-15 minute intervals (due to the lag described above).