Here are a few potential design patterns you could try. The first puts a timer catch event on each user task and after a set period of time, interrupts the user task and sends it on to the parallel gateway. Both the primary outflow for each user task and the outflow coming out of the timer boundary event merge together, so if all user tasks are completed before the allotted time, the instance will move on, otherwise the boundary events will trigger and push everything to the parallel gateway to move on. 

The second pattern is similar but uses a timer boundary event on a sub-process and has all the user tasks in the sub-process. This might save you some configuration time (just one timer boundary event) but you couldn't have different time limits for each user task. If you're ok with the same time limit for each user task, you can use method 2 below. (You don't need the incoming parallel gateway, just draw the arrow from your previous step). The sub-process will be activated and inside, will split into each user task. As they are finished, they wait at the parallel gateway inside the sub-process. If they are all completed, it finishes the sub-process and moves on to the subsequent user task. If the timer boundary runs out and interrupts the sub-process, it will end all the activity in the sub-process wherever it's at and move on to the summary. Anyone who finished on time has their answers submitted, it ignores anyone that didn't submit.