LTE is increasingly seen as a system for serving real-time Machine-to-Machine (M2M) communication needs. The asynchronous M2M user access in LTE is obtained through a two-phase access reservation protocol (contention and data phase). Existing analysis related to these protocols is based on the following assumptions: (1) there are sufficient resources in the data phase for all detected contention tokens, and (2) the base station is able to detect collisions, i.e., tokens activated by multiple users. These assumptions are not always applicable to LTE - specifically, (1) due to the variable amount of available data resources caused by variable load, and (2) detection of collisions in contention phase may not be possible. All of this affects transmission of real-time M2M traffic, where data packets have to be sent within a deadline and may have only one contention opportunity. We analyze the features of the two-phase LTE reservation protocol and asses its performance, when assumptions (1) and (2) do not hold.